Adap%ve Forwarding Strategy for Hyperbolic Rou%ng · viable rou%ng protocol • Number of rou%ng...
Transcript of Adap%ve Forwarding Strategy for Hyperbolic Rou%ng · viable rou%ng protocol • Number of rou%ng...
Adap%veForwardingStrategyforHyperbolicRou%ng
VinceLehman,AshleshGawande,LanWang,UniversityofMemphis
RodrigoAldecoa,DmitriKrioukov,NortheasternUniversityBeichuanZhang,UniversityofArizona
LixiaZhang,UCLA
1
Rou%ngScalabilityinNDN
• Alreadylargenumberofcontentnamesintoday’sInternet
• NumberofForwardingInforma%onBase(FIB)entriesinNDNcouldgrowatanunmanageableratewithoutviablerou%ngprotocol
• Numberofrou%ngupdates(overhead)tomaintainconsistentFIBsmayalsobecostly
• NDNnetworksmustscaleintermsofrou%ngtablesizeandrou%ngprotocoloverhead
2
Whatdowewantfromarou%ngprotocol?
• SmallFIBsize• Lownumberofrou%ngupdates• Comparableperformancetoshortestpathrou%ngalgorithms
• Inshort,wewanttoboundthesizeofrou%ngstatewhilesuppor%nganunboundednamespace
3
Poten%alofAdap%veForwarding
• Rou%ngcanbelessdynamicbecauseofadap%veforwardingplane
• NDNForwardingStrategycanmakeforwardingdecisionsbasedonrou%ng
• Non-tradi%onalrou%ngschemesmaybecomeviable
4
HyperbolicRou%ng(HR)• Eachnodeandnameprefixhasasetofhyperboliccoordinates
• Nexthopranksarecalculatedbasedoneachneighbor’sdistancetothedes%na%on
• Noneedtodistributetopologyinforma%onorupdates
• ForforwardinginHR,itisnecessarythat:1. Eachnodeknowsitsownhyperboliccoordinates2. Eachnodeknowsitsneighbors’coordinates3. Thenodethatisforwardingknowsthecoordinates
ofthedes%na%on
5
GreedyForwardinginHR
Toforwardapacket:– Findtheneighborclosesttothedes%na%on– Forwardthepackettothatneighbor
Des%na%on Nexthops
D {A,cost=10},{B,cost=30}
6
Neighbor’scoordinates(θ,r)Des%na%oncoordinates(θʹ,rʹ)∆θ=|θ−θʹ|modπdistance=acosh(coshrcoshrʹ−sinhrsinhrʹcos∆θ)
WhyHyperbolicRou%ng?
• Intheidealcase,noFIBisneededIfanInterestcarriesthecoordinatesassociatedwiththename,eachnodeonlyneedstoknowtheirneighbors’coordinates
• Lowcommunica%oncostFewrou%ngupdates,ascoordinatesrarelychange
7
DrawbacksofHR
• Isthereaddi%onaldelay?
• Ispacketforwardingperformancecomparabletoashortestpathprotocolsuchaslink-state?
• Canpacketsbecome“lost”inHRwithgreedyforwarding?
8
Sub-Op%malPathsinHR
Toforwardapacket:– Findtheneighborclosesttothedes%na%on– Forwardthepackettothatneighbor
Des%na%on Nexthops
D {B,cost=10},{A,cost=20}
5ms 5ms
10ms 10ms
9
LocalMinimainHR• Occurswhenanodedoes
nothaveaneighborclosertothedes%na%oncoordinatesthanitself
• ForwardinginHRcanhitadead-end
• NDNcanusemul%plenexthopsperprefixandadap%veforwardingtoavoidthisproblem
• Numberofnexthopsperprefixcanbelimitedtoreduceoverhead(mul%pathfactor)
S
B
D
A
Localminima(ClosesttoD;noneighborcloser)
10
HRwithBest-RouteStrategy• BestRouteStrategysimplyusesthenexthoprankedhighest
bytherou%ngprotocol• Delaystretch–Packetdelayra%oofRTTinHRoverRTTin
shortestpathrou%ng
11
Highdelaystretch
ForwardingStrategyDesign• Goal:Findanduseop%malpaths• Approaches:
– TestdifferentpathsbyperiodicallyforwardingInteresttodifferentFaceusingnewnonce
– MaintainSRTTmeasurementforeachFacetodeterminebestFaceforforwarding
– Probabilis%callychooseFacestoprobe;weightedtowardsbeqerperformingFaces
12
Adap%veSRTT-BasedForwarding
• ConsiderRoundTripTime(RTT)whenchoosingnexthopinHR– UseSmoothedRTT(SRTT)toallowvaria%on
• BestSRTT-BasedForwarding– ChoosenexthopforeachFIBentrybasedonSRTT
• Probabilis%cSRTT-BasedProbing– PeriodicallyprobeunusednexthopstolearnRTT– Nexthopsthatperformedwellpreviouslyhavehigherprobability
13
ChoosingBestFaceforForwarding
• ForaFIBentry:– SortavailableFacesintogroups:
1. FaceswithSRTTmeasurementsandno%meoutsaresortedbySRTT;alowerSRTTisbeqer
2. FaceswithoutSRTTmeasurementsandno%meoutsaresortedbyrou%ngcost
3. FaceswithSRTTmeasurementsandtheirlastInterest%med-outaresortedbyrou%ngcost
– ChoosethefirstFacefrom1,else2,else3
14
Probabilis%cSRTT-BasedProbing• WanttolearnRTTofnexthopsotherthantheonebeing
usedforforwarding
• SchedulefirstprobeforaFIBentryonfirstmatchingInterestininterval[0,T1]
• WhenanInterestisforwardedthatmatchesaFIBentryandtheFIBentryisdueforprobing:– PickFacetoprobeandforwardprobeInterestwithnewnonce
• Auereachprobe,nextprobeisscheduledinT2seconds
15
Probabilis%cSRTT-BasedProbing
• Selec%nganexthoptoprobe:– Usesamegroupingandsor%ngfornexthopsasinforwarding(excludeprimarynexthop)
– Choosethelowestcostnexthopwithnomeasurements(Group2)immediatelyifoneexists
– Otherwise,combineGroup1andGroup3andassigneachnexthopaprobingprobability
– Nexthopsthatperformedwellpreviouslyareassignedahigherprobability
16
Probabilis%cSRTT-BasedProbing
• Computeprobability:– N=Numberofnexthopsincombinedlist– X=Sumofindicesincombinedlist;n(n+1)/2– P(Nexthopi)=(N+1-i)/X;
• Selectnexthoptoprobebasedonprobability
17
ProbingProbabilityExample
18
NACKs• AllIncomingNACKsarehandledbyseyngthe%meoutflagin
thecorrespondingface'smeasurements
• NACKsdonotcontributetoSRTTcalcula%on
• Ifstrategycannotforwardtoanynexthops,sendNO_ROUTENACK
19
Retransmissions
• Consumerretransmissions– SuppressinsamemannerasBestRouteStrategy2– Interestissuppressedifwithinminimuminterval– Otherwise,pickbestnexthoptoforwardtobasedonsortedgroups
• Routerretransmissions– Routerretransmissionsarenotperformed
20
HRwithASFStrategy
21